Crush v0.94.1: Bring Your Own ChatGPT Subscription
Headless users get a new `--reasoning-effort` flag in `crush run` to control model reasoning depth.
This release lets you use your own ChatGPT subscription inside Crush, giving direct control over costs and access. The non-interactive `--reasoning-effort` flag adds fine-grained reasoning control for automated pipelines, making Crush more flexible for scripted or server-side workflows without a GUI.
Source: GitHub
Headless users get a new `--reasoning-effort` flag in `crush run` to control model reasoning depth.
This release lets you use your own ChatGPT subscription inside Crush, giving direct control over costs and access. The non-interactive `--reasoning-effort` flag adds fine-grained reasoning control for automated pipelines, making Crush more flexible for scripted or server-side workflows without a GUI.
Source: GitHub
MAI-Code-1-Flash deprecated in GitHub Copilot
GitHub has deprecated MAI-Code-1-Flash across all Copilot experiences, including Chat, inline edits, ask/agent modes, and code completions.
If you use MAI-Code-1-Flash, update workflows to a supported alternative. Enterprise admins may need to enable the new model in Copilot settings. No action required to remove the deprecated model.
Source: GitHub Copilot
GitHub has deprecated MAI-Code-1-Flash across all Copilot experiences, including Chat, inline edits, ask/agent modes, and code completions.
If you use MAI-Code-1-Flash, update workflows to a supported alternative. Enterprise admins may need to enable the new model in Copilot settings. No action required to remove the deprecated model.
Source: GitHub Copilot
Google Cloud releases a plugin for AI coding agents
Installable bundles that give your agent Google Cloud skills and tools.
Developers can now equip any AI coding agent with direct Google Cloud capabilities, making cloud workflows more efficient from within the editor.
Source: google.com
Installable bundles that give your agent Google Cloud skills and tools.
Developers can now equip any AI coding agent with direct Google Cloud capabilities, making cloud workflows more efficient from within the editor.
Source: google.com
Cline desktop v0.0.27: cleaner sign-in, OpenCode Go fix, Crusoe provider
An expired Cline sign-in now shows one actionable error instead of two dead ends. OpenCode Go models work again.
Developers get a clearer sign-in failure path, restored OpenCode Go compatibility, and a new Crusoe provider option for OpenAI-compatible workflows.
Source: GitHub
An expired Cline sign-in now shows one actionable error instead of two dead ends. OpenCode Go models work again.
Developers get a clearer sign-in failure path, restored OpenCode Go compatibility, and a new Crusoe provider option for OpenAI-compatible workflows.
Source: GitHub
Copilot code review auto-resolves comments and writes smarter commit messages
Push a fix, and Copilot closes the thread. Apply a suggestion, and it drafts the commit message.
Less manual cleanup, clearer signal on what still needs your attention, and deeper validation behind the scenes.
Source: GitHub Copilot
Push a fix, and Copilot closes the thread. Apply a suggestion, and it drafts the commit message.
Less manual cleanup, clearer signal on what still needs your attention, and deeper validation behind the scenes.
Source: GitHub Copilot
Agentic Batch Changes: One Prompt, Thousands of Repos
Sourcegraph launches Agentic Batch Changes for Cloud users, letting one engineer describe a codebase-wide change in plain language and have an agent scope, adapt, and roll it out across hundreds or thousands of repos.
Coding agents make single-repo changes fast but leave the rest of the codebase drifting. Agentic Batch Changes orchestrates agents or scripts per repo, adapts to differences, retries on CI failure, and tracks merge status—so a large migration no longer means a spreadsheet or a hundred manual PRs.
Source: Sourcegraph
Sourcegraph launches Agentic Batch Changes for Cloud users, letting one engineer describe a codebase-wide change in plain language and have an agent scope, adapt, and roll it out across hundreds or thousands of repos.
Coding agents make single-repo changes fast but leave the rest of the codebase drifting. Agentic Batch Changes orchestrates agents or scripts per repo, adapts to differences, retries on CI failure, and tracks merge status—so a large migration no longer means a spreadsheet or a hundred manual PRs.
Source: Sourcegraph
Qwen Code v0.23.4: Auto-start Goals, Peer SDK, and Turn Limits
Approved Web Shell Goals now launch automatically after their turn ends, eliminating manual /goal set commands.
Developers get faster workflows with auto-starting Goals, a new peer endpoint for cross-session messaging, and optional turn/active-time limits via model.goalMaxTurns and model.goalMaxActiveMinutes. Active Goal time also resets on restore.
Source: GitHub
Approved Web Shell Goals now launch automatically after their turn ends, eliminating manual /goal set commands.
Developers get faster workflows with auto-starting Goals, a new peer endpoint for cross-session messaging, and optional turn/active-time limits via model.goalMaxTurns and model.goalMaxActiveMinutes. Active Goal time also resets on restore.
Source: GitHub
Claude Code Now Runs on Your Own Infrastructure via Coder Agent Relay
Early access for select design partners lets regulated enterprises run Claude Code inside self-hosted, network-isolated workspaces while Anthropic handles the agent loop.
Financial services, government, and other compliance-heavy orgs can finally adopt Claude Code without ceding control of code, credentials, or egress. Platform teams reuse existing workspace policies and audit trails—no fresh security review needed.
Source: Coder
Early access for select design partners lets regulated enterprises run Claude Code inside self-hosted, network-isolated workspaces while Anthropic handles the agent loop.
Financial services, government, and other compliance-heavy orgs can finally adopt Claude Code without ceding control of code, credentials, or egress. Platform teams reuse existing workspace policies and audit trails—no fresh security review needed.
Source: Coder
TermSquad launches managed cloud computer for AI coding agents
Developers can now run persistent Linux sessions for AI coding agents across devices, with multi-agent orchestration and shared project memory.
This eliminates the need to manage your own cloud VM for agent workloads. TermSquad handles the infrastructure, letting you focus on coding with tools like Codex, Claude Code, Cursor, or Copilot.
Source: macaubusiness.com
Developers can now run persistent Linux sessions for AI coding agents across devices, with multi-agent orchestration and shared project memory.
This eliminates the need to manage your own cloud VM for agent workloads. TermSquad handles the infrastructure, letting you focus on coding with tools like Codex, Claude Code, Cursor, or Copilot.
Source: macaubusiness.com
Unity Plugin for Codex Now Available
OpenAI's Codex gets an official Unity plugin, compatible with Unity 6+ and offering 31 initial skills written by Unity's own feature teams.
Setup takes two terminal commands or one click from the OpenAI plugin directory. The unity-cli skill handles Editor installs, project creation, and package management, letting developers stay in the coding flow without switching tools.
Source: BigGo Finance
OpenAI's Codex gets an official Unity plugin, compatible with Unity 6+ and offering 31 initial skills written by Unity's own feature teams.
Setup takes two terminal commands or one click from the OpenAI plugin directory. The unity-cli skill handles Editor installs, project creation, and package management, letting developers stay in the coding flow without switching tools.
Source: BigGo Finance
Goose v1.51.0 adds EUrouter provider, GPT-live API, and gateway pairing allowlist
Goose v1.51.0 is out, tagged on main and published 2026-09-17. The release notes call out EUrouter as a declarative provider, GPT-live API support, and an operator allowlist for gateway pairing.
If you run Goose with EUrouter or GPT-live, or pair gateways, this release changes what you can configure. The allowlist gives operators control over which gateways can pair; check the notes before upgrading.
Source: GitHub
Goose v1.51.0 is out, tagged on main and published 2026-09-17. The release notes call out EUrouter as a declarative provider, GPT-live API support, and an operator allowlist for gateway pairing.
If you run Goose with EUrouter or GPT-live, or pair gateways, this release changes what you can configure. The allowlist gives operators control over which gateways can pair; check the notes before upgrading.
Source: GitHub
Copilot CLI agentic metrics land in the usage metrics API
GitHub Copilot's usage metrics API now reports CLI agentic activity for skills, custom agents, MCP servers, slash commands, and plugins, across enterprise and organization reports.
Admins can see which CLI customizations gain traction and where enablement gaps are. Customer-defined names stay hidden, and plugin totals are a subset of skill totals, so don't add them.
Source: GitHub Copilot
GitHub Copilot's usage metrics API now reports CLI agentic activity for skills, custom agents, MCP servers, slash commands, and plugins, across enterprise and organization reports.
Admins can see which CLI customizations gain traction and where enablement gaps are. Customer-defined names stay hidden, and plugin totals are a subset of skill totals, so don't add them.
Source: GitHub Copilot
🔥1
What this channel is
AI coding tools and agent harnesses for daily work: what changed in a release, whether it is worth updating, and what it changes in the way you work.
Every post links its primary source and skips the "new version released" padding: no changelog dumps, no benchmark claims without context.
Sibling channels: @llm_ai_for_devs — model releases and API pricing, @ai_dev_security — CVEs and dependency risk.
AI coding tools and agent harnesses for daily work: what changed in a release, whether it is worth updating, and what it changes in the way you work.
Every post links its primary source and skips the "new version released" padding: no changelog dumps, no benchmark claims without context.
Sibling channels: @llm_ai_for_devs — model releases and API pricing, @ai_dev_security — CVEs and dependency risk.
Coding Agents — AI dev tools & harnesses pinned «What this channel is AI coding tools and agent harnesses for daily work: what changed in a release, whether it is worth updating, and what it changes in the way you work. Every post links its primary source and skips the "new version released" padding: no…»
GitHub Copilot deprecates six models on October 19th, 2026
GitHub will deprecate Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5 across all Copilot experiences, including Chat, inline edits, ask and agent modes, and code completions.
Update workflows and integrations to the suggested alternatives before the date. Default enablement covers Copilot Enterprise and Business unless an admin disabled it; otherwise enable models via Copilot settings. No removal action is needed.
Source: GitHub Copilot
GitHub will deprecate Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5 across all Copilot experiences, including Chat, inline edits, ask and agent modes, and code completions.
Update workflows and integrations to the suggested alternatives before the date. Default enablement covers Copilot Enterprise and Business unless an admin disabled it; otherwise enable models via Copilot settings. No removal action is needed.
Source: GitHub Copilot
Copilot Code Review Adds Progress Tracking, Smarter Auto-Resolve, and Batch Commit Messages
Copilot code review now shows how a review changes over time, auto-resolves its own comments more intelligently, and writes commit messages for eligible suggestion batches. GitHub says these updates are generally available.
Findings are grouped into Open, Resolved since last review, and Previously missed, each with severity and an inline comment link. Copilot resolves comments with a reason, either Won't Fix or Incorrect, based on later commits.
Source: GitHub Copilot
Copilot code review now shows how a review changes over time, auto-resolves its own comments more intelligently, and writes commit messages for eligible suggestion batches. GitHub says these updates are generally available.
Findings are grouped into Open, Resolved since last review, and Previously missed, each with severity and an inline comment link. Copilot resolves comments with a reason, either Won't Fix or Incorrect, based on later commits.
Source: GitHub Copilot
AWS Strands harness ships a general-purpose agent harness for Python and TypeScript
The Strands Agents team at AWS released Strands harness, a fully assembled general-purpose agent harness. It runs locally or deploys to a cloud provider, ships for Python and TypeScript under Apache 2.0, and starts with one line of code.
Install with pip install strands-harness or npm install @strands-agents/harness. It ships shell, file, and web tools, keeps long-term memory across runs, resumes from a session ID, and loads Agent Skills. The team reports 28% lower cost than other harnesses across 6 benchmarks.
Source: MarkTechPost
The Strands Agents team at AWS released Strands harness, a fully assembled general-purpose agent harness. It runs locally or deploys to a cloud provider, ships for Python and TypeScript under Apache 2.0, and starts with one line of code.
Install with pip install strands-harness or npm install @strands-agents/harness. It ships shell, file, and web tools, keeps long-term memory across runs, resumes from a session ID, and loads Agent Skills. The team reports 28% lower cost than other harnesses across 6 benchmarks.
Source: MarkTechPost
Grok 4.7 ships with a 500,000-token context window
SpaceXAI released Grok 4.7, a larger base model than Grok 4.6, available today in Cursor and Grok Build, plus the Grok API, third-party coding harnesses, and model routers and cloud platforms.
API model id is grok-4.7. Pricing starts at the same rates as Grok 4.6: $2 per million input tokens and $6 per million output tokens below 200,000 tokens of context. It is trained to natively understand the Grok Bot harness.
Source: x.ai
SpaceXAI released Grok 4.7, a larger base model than Grok 4.6, available today in Cursor and Grok Build, plus the Grok API, third-party coding harnesses, and model routers and cloud platforms.
API model id is grok-4.7. Pricing starts at the same rates as Grok 4.6: $2 per million input tokens and $6 per million output tokens below 200,000 tokens of context. It is trained to natively understand the Grok Bot harness.
Source: x.ai
❤1
GitHub Copilot app adds OpenTelemetry via enterprise-managed settings
The GitHub Copilot app now supports OpenTelemetry configuration through enterprise-managed settings, letting administrators send agent activity data to their organization's compatible monitoring tools.
Admins configure the telemetry property in managed-settings.json to enable export and set the receiving endpoint. Prompt and response content is excluded by default, so review content-capture settings before enabling it.
Source: GitHub Copilot
The GitHub Copilot app now supports OpenTelemetry configuration through enterprise-managed settings, letting administrators send agent activity data to their organization's compatible monitoring tools.
Admins configure the telemetry property in managed-settings.json to enable export and set the receiving endpoint. Prompt and response content is excluded by default, so review content-capture settings before enabling it.
Source: GitHub Copilot
NVIDIA, NTU and MIT release SoL-Pi, 4 efficiency mechanisms for the Pi coding agent
SoL-Pi ships on GitHub under NVlabs as an MIT-licensed extension that runs on an unmodified Pi release. It is tested with Pi 0.85.1 and Node.js 22.19 or newer.
On the 51-task EdgeBench evaluation, SoL-Pi cuts recorded token traffic by 44.7% to 49.0% versus Pi and cuts API cost by roughly 33%. Action Fusion merges both into 1 tool request and returns both outcomes in 1 observation; ObservationPack archives tool outputs above 10 KiB locally and sends them in full for the next 2 provider requests.
Source: MarkTechPost
SoL-Pi ships on GitHub under NVlabs as an MIT-licensed extension that runs on an unmodified Pi release. It is tested with Pi 0.85.1 and Node.js 22.19 or newer.
On the 51-task EdgeBench evaluation, SoL-Pi cuts recorded token traffic by 44.7% to 49.0% versus Pi and cuts API cost by roughly 33%. Action Fusion merges both into 1 tool request and returns both outcomes in 1 observation; ObservationPack archives tool outputs above 10 KiB locally and sends them in full for the next 2 provider requests.
Source: MarkTechPost
👍1